Skip to content
The category

Spec Fidelity

How faithfully an implementation matches its specification, expressed as one reproducible number. Tests, linters, scanners, and review each measure code against an external standard. Spec Fidelity is the only one that measures it against your spec.

A

Why it is not testing, linting, QA, or review

Tests

Assert behavior you wrote

A test encodes what a developer expected. If the developer misread the spec, the test passes and the requirement still fails. Tests cannot catch a faithful-looking misread.

Linters

Enforce style, not meaning

Linters keep code consistent and safe from a class of bugs. They have no model of what the feature was supposed to do.

QA

Samples behavior late

Manual and automated QA find issues after the build exists, against test plans that are themselves a lossy copy of the spec.

Code review

Judges craft, under time pressure

Reviewers check quality and taste on a diff. Few re-read the full requirement for every line, and AI diffs are large and plausible.

Security scanners

Match known risk patterns

Scanners look for vulnerability signatures. A correct-but-wrong feature has no signature to match.

Spec Fidelity

Measures faithfulness to intent

Norma reads the requirement and the implementation and judges whether they agree. That is the gap every other tool leaves open.

B

Why existing tools miss this

ToolWhat it checks
TestsBehavior
LintersStyle
Code reviewCode quality
Security scannersVulnerabilities
NormaRequirement fidelity

Every other tool measures your code against an external standard. Norma is the only one that measures it against your spec. That gap is the whole reason it exists.

C

Introducing the Spec Fidelity Score

A single number for how faithfully an implementation matches its specification. Read it like Lighthouse reads performance. Spec Fidelity is the category, and Norma is the tool that measures it.

95+
Ship
Faithful to the spec. Merge with confidence.
80 to 94
Review
Mostly there. A human should look at the gaps before release.
below 80
Investigate
Material drift, or a contradiction. Do not ship on trust.

Every Spec Fidelity Score should stamp the model and prompt version, so it is reproducible and contestable. The report has an open schema, which is how the number becomes a shared standard rather than one vendor's metric.

Did the build keep its promises?

AI can write the code.
Who verifies the intent?